A nonparametric test for interaction between two treatments which had been previously reported in the literature was found to be seriously affected by the presence of first-order effects in the data. A modification of the test was developed that mitigated this difficulty. The modified test was shown by a Monte-Carlo study to behave quite favorably relative to other nonparametric tests for interaction.